Looking to find out about Google Gemini AI? Then you have come to the right place. Google’s latest foray into the AI space has garnered widespread media attention after the company posted results that appeared to show their model outperformed OpenAI’s ChatGPT. According to Google DeepMind’s CEO, Demis Hassabis, the new artificial intelligence will be their most advanced LLM (large language models) architecture yet.

“For a long time, we've wanted to build a new generation of AI models, inspired by the way people understand and interact with the world,” writes Hassabis on the Google blog. “AI that feels less like a smart piece of software and more like something useful and intuitive — an expert helper or assistant. Today, we're a step closer to this vision as we introduce Gemini, the most capable and general model we've ever built.”

So, what does that mean for consumers in practical terms? Here’s everything you need to know about Google’s Gemini AI, and what it could mean for the main player in town right now, Open AI.

What is Google Gemini AI?

Gemini was announced during the Google I/O conference last year. Gemini AI is Google’s next-generation foundation model and is a large language model (LLM) similar to GPT-4. According to Hassabis, Gemini is DeepMind’s most ‘flexible model yet’, capable to run on mobile devices. The first version of Gemini is version 1.0 – and this will come in three different sizes.

• Gemini Ultra — the largest and most capable model for highly complex tasks.
• Gemini Pro — the best model for scaling across a wide range of tasks.
• Gemini Nano — the most efficient model for on-device tasks

So, what can Gemini be used for? As alluded to previously, the team at DeepMind has been putting the AI through its paces in a series of natural image, audio and video tests as as well as mathematical reasoning. They claim that Gemini exceeds 30/32 LLM benchmarks used to assess state-of-the-art performance. Furthermore, Gemini also outperformed human experts on MMLU ((massive multitask language understanding) with a score of 90%. Gemini even managed a state-of-the-art score of 59.4% on a new benchmark called MMMU – which tests deliberate reasoning.

This is no surprise given that Google designed Gemini to understand and reason about inputs from the ground up. Thanks to these enhanced reasoning abilities, Gemini can sift through a sea of data to uncover what you’re looking to find. This could have huge implications in a number of practical tasks – for example, trying to find particular examples to support what you have written in an essay. It also means Gemini is excellent at maths and physics-based tasks. Google trained the model to understand text, images, audio, ‘and more’ at the same time.

“At a high level you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models.”

Demis Hassabis, CEO of Google DeepMind

What is AlphaCode 2?

Gemini, as is the case with many AI models, can also create high-quality code in programming languages including Python, Java, C++, and Go. DeepMind claims its ability to work across languages and reason about complex information “makes it one of the leading foundation models for coding in the world”. There’s also a new version of AlphaCode called AlphaCode 2, which builds on the former’s formidable reputation to offer improved performance when evaluated against the original’s test parameters.

Elsewhere, it’s worth noting that Gemini has the ‘most comprehensive safety evaluations’ of any Google AI model to date, including for bias and toxicity. This means the model has been stringently tested to mitigate risks from areas such as cyber-offense, persuasion and autonomy.

How to access Gemini AI

Gemini AI is still in development and there are processes to be completed so it is not yet possible to access it. The company still has work to do and it could cost tens or hundreds of millions of dollars to complete. GPT-4 cost more than $100 million to create. That being said, on December developers and enterprise customers can access Gemini Pro via the Gemini API in Google AI Studio or Google Cloud Vertex AI.

“Android developers will also be able to build with Gemini Nano, our most efficient model for on-device tasks, via AICore, a new system capability available in Android 14, starting on Pixel 8 Pro devices,” reads the Google blog. You can sign up for an early preview of AICore here. Gemini Ultra, meanwhile, won’t be available until later this year, as it’s currently going through a number of pre-launch tests.

If you’re keen to give it a go, check out how to access Google Gemini with our trusty step-by-step guide, which walks you through the simple process.

Essential AI Tools

Will Gemini AI be more powerful than ChatGPT? Our opinion

Google’s foray into artificial intelligence seemed at first to lag behind the likes of OpenAI. However, recent updates on Gemini AI indicate it will operate to an exceedingly high level, even more so than its close competitors. A recent report by SemiAnalysis indicated Google’s AI system is five times more powerful than other top AI services like ChatGPT.

Furthermore, it suggested that by 2024, this year, Google’s Gemini model may be as much as 20 times more powerful. The early results highlighted in the Google blog do seem to support the notion that Gemini is indeed more powerful.

Sundar Pichai, CEO of Google and Alphabet, is confident in his ability to keep the tech giant ahead in the AI arms race. He explains that “Gemini was created from the ground up to be multimodal, highly efficient at tool and API integrations and built to enable future innovations, like memory and planning,” also noting that “While still early, we're already seeing impressive multimodal capabilities not seen in prior models.” We’ll know more when the Ultra version lands in the coming weeks or months.

FAQ

Who owns DeepMind?

DeepMind is a Google-owned company. The machine learning AI startup was co-founded in 2010 by Demis Hassabis, Shane Legg, and Mustafa Suleyman who has since joined Reid Hoffman on Inflection AI. The company, based in Kings Cross, London, became a subsidiary of Google following the acquisition in 2014.

Is Google Gemini AI better than ChatGPT?

Google Gemini is multimodal – that is, able to natively receive inputs of more than one type of media (text, image, audio, and video) – with a core focus on image generation. This means it competes with OpenAI’s GPT-4V. Across over a dozen different benchmarks, Gemini outperforms the ChatGPT model in most tests.